Multiple StellarPopulation instances

Because parameter state and dirtiness is specific to each instance, but the computed model is held by the module level driver object, having multiple StellarPopulation instances can cause a model to not be recomputed even though it should be (example code below) This might also cause problems with shared memory multiprocessing.

Some documentation advising only a single sps instance per program might be a good idea.

from numpy.testing import assert_allclose
import fsps
sps = fsps.StellarPopulation(zcontinuous=1, vactoair_flag=False)
pop = fsps.StellarPopulation(zcontinuous=1, vactoair_flag=False)
w1, s1 = sps.get_spectrum(tage=10.0)
w, s = pop.get_spectrum(tage=1.0)
w2, s2 = sps.get_spectrum(tage=10.0)
assert_allclose(s, s2) # this should not pass, but does
assert_allclose(s1, s2) # this should not fail, but does

Comienza con StellarPopulation.get_spectrum y el estado del controlador a nivel de módulo descrito en el issue. Reproduce la secuencia proporcionada con dos instancias y, a continuación, rastrea cómo se comparten la condición de modificación de los parámetros y el recálculo del modelo. Se considera terminado cuando cada instancia devuelve resultados coherentes con sus propios parámetros, incluidas las dos assertions del ejemplo.